How we measure — and what the number means.

We ask AI what a real patient would ask.

For each city we run roughly 1000+ measurement queries — in Peterborough, 1035 answers — phrased the way a prospective patient actually asks. We record every practice named in every response, then aggregate.

Six AI surfaces, measured the same way.

The same queries run across the six surfaces patients actually use. We don't cherry-pick a favourable engine — every practice is measured against all six, in the same week.

Mention rate, with an honest denominator.

Your mention rate is simple: of all the answers we collected on a surface, what share named your practice. The denominator is the total number of runs on that surface — not just the answers you happened to appear in. Runs where no local practice was named still count, because being absent from those answers is exactly the thing we're measuring.

Google AI Overviews is measured only on searches where an AI Overview actually appeared — the honest denominator for that surface.

Integer rates, and a stated noise floor.

Repeated measurements across the panel establish a noise floor of about ±6 percentage points at the aggregate level. Because the data doesn't support more precision than that, we report whole-number rates only — no false decimals, no fabricated exactness.

This is deliberate. A number that looks more precise than the measurement behind it is a score dressed up as a fact. We'd rather give you a rate you can trust than a decimal you can't.

AI answers shift, so we remeasure.

AI systems change their answers over time as models and indexes update. A measurement is a snapshot of a moment, and we say so. Reports carry the date they were measured and a suggested remeasurement window, so you're never reading a stale number as if it were today's.